Amazon and Google are already armed

Let’s go back in time for a moment. By mid-2023, ChatGPT He had been between us for half a year and artificial intelligence was experiencing a huge boom. For some it was a fad; For others, one of the most relevant technological disruptions of recent times.

OpenAI had the lead. From being practically unknown, she went on to dominate headlines. He achieved this by putting a product as fascinating as it was imperfect in the hands of the public, the type of launch that Big Tech would hardly have materialized.

In its early months, ChatGPT had few guardrails and made mistakes quite frequently. Such a setback would have meant a reputational blow Huge for any tech giant, but a startup could possibly take that risk.

NVIDIA, the winner in the AI ​​race, faces a challenge

The basis for competing head-to-head with ChatGPT was—and continues to be—developing increasingly advanced language models. After launching GPT-3.5the model that gave rise to ChatGPT, OpenAI moved quickly to introduce GPT-4 in March 2023.

From that moment on, the old and new players in the sector had no alternative but to enter the game, unless they wanted to stay out of what was already aiming to be the next technological revolution. And whatever the name of the company, they all depended on one key player: NVIDIA.

The reason? Jensen Huang’s firm had the best AI-specialized GPUs on the market, with the H100 as standard bearer. And we talk about GPUs because they are much more suitable than CPUs for the parallel processing tasks that modern AI requires.

The approach was simple: if you wanted to compete in generative AI—and hopefully aspire to lead it—you needed to buy NVIDIA GPUs, upgrade your data centers or open new ones. And all this against the clock. The result was massive demand that sometimes led to shortages.

An illustrative example: at the beginning of 2024, Meta announced that Its renewed infrastructure for generative AI would reach 350,000 NVIDIA H100 GPUs by the end of the year, with a power equivalent to about 600,000 H100. By June 2024, NVIDIA was already the most valuable listed company on the planet.

It is often said that the difficult thing is not to get to the top—that too—but to stay there. That’s exactly the challenge NVIDIA faces now. Their products remain at the top and its strategy covers much more than hardware: it includes a stack of software designed to make the most of the architecture CUDA.

But the competition is knocking on the door loudly, the same as with OpenAI. The leadership of these years is pressured by increasingly fierce and global competition.

Amazon and Google’s bet on AI chips

The most recent move comes from Amazon, which has just introduced Trainium3 UltraServer, a system powered by its AI chip Trainium3 manufactured in 3 nanometers. According to the companythese chips and systems are 40% more energy efficient than the previous generation.

And it doesn’t end there. Amazon also showed its roadmap for Trainium4, already in development. It promised support for NVLink Fusion, NVIDIA’s high-speed chip interconnect technology, opening the door to interoperable systems.

Google, which has been developing its TPU chips for AI for a decade, is starting to bet on its own hardware to boost Gemini. Even rumors circulate that talk about a multimillion-dollar investment by Meta to buy AI chips from Google, a move that enters squarely into AI territory.

We would also be facing a quite notable change of course in the search engine company, which until now had limited the use of its chips to its own data centers.

Do you remember that Meta is one of NVIDIA’s big clients? Well, it is also developing—and testing—its own AI chips, such as the MTIA. Even OpenAI has decided to partner with Broadcom to design their own hardware for their data centers.

In China, the scenario does not favor NVIDIA either. The trade war with the United States has narrowed its room for maneuver in the country, while local actors such as Huawei they push their own chipswith the Ascend 910D already in production and Ascend 920 new generation.

Ultimately, we are seeing many of the players who turned to NVIDIA in the early stages of the AI ​​race beginning to follow their own path. The reasons are multiple, but one stands out above all: the need for independence in a technological competition that only intensifies.

Images | World Governments Summit | amazon | Google

In Xataka | If you think the internet was much better before AI, congratulations: they have created an extension for you